enh(array, iop): add new Array attributes and dunders, and corresponding iop functions

[nicola-bastianello]

this PR adds the following dunders to Array:

• arithmetic operations: __pos__, __floordiv__, __mod__
• bitwise operations: __invert__, __or__, __xor__, __lshift__, __rshift__
• in-place: __ifloordiv__, __ipow__, __imod__, __imatmul__, __iand__, __ior__, __ixor__, __ilshift__, __irshift__
• reflected: __rfloordiv__, __rpow__, __rmod__, __ror__, __rxor__, __rlshift__, __rrshift__
• methods: __bool__, __index__, __int__

it also adds the Array attributes:

• dtype, mT (matrix transpose)

finally, it adds the iop functions corresponding to the new dunders/attributes:
for iop:

• element-wise functions: positive, remainder, bitwise_left_shift, bitwise_invert, bitwise_or, bitwise_right_shift, bitwise_xor, floor_divide
• linear algebra: matrix_transpose

tests for all new features were also added

[Copilot]

Pull request overview

This PR expands decent_array.Array and the decent_array.interoperability (“iop”) surface area to support additional unary/binary/in-place/reflected operators, plus new Array attributes (dtype, mT) and corresponding backend/iop functions (notably matrix_transpose, floor_divide, remainder, and additional bitwise ops).

Changes:

• Added new Array dunder operators for floor-division/modulo, unary plus, additional bitwise ops/shifts, and scalar coercions (__bool__, __int__, __index__).
• Added iop functions for matrix_transpose, floor_divide, remainder, and bitwise operators, with backend implementations across NumPy/JAX/PyTorch/TensorFlow.
• Added tests covering the new iop functions and Array behaviors (including mT and additional operators).

Reviewed changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
tests/test_iop_functions.py	Adds tests for matrix_transpose, floor_divide, remainder, and new backend in-place ops + bitwise iop functions.
tests/test_array.py	Adds tests for new Array operators (//, %, bitwise ops/shifts) and mT behavior/validation.
decent_array/types.py	Introduces _STRING_TO_DTYPE lookup map for DTypes.
decent_array/_array.py	Implements new dunders, adds dtype and mT properties, and updates operator hot-path notes.
decent_array/interoperability/_abstracts/backend.py	Extends the backend interface with matrix_transpose, new in-place ops, floor-div/mod, and bitwise ops/shifts.
decent_array/interoperability/_iop/math.py	Adds floor_divide, remainder, and positive iop functions.
decent_array/interoperability/_iop/manipulations.py	Adds matrix_transpose iop function.
decent_array/interoperability/_iop/bit_operators.py	Adds iop wrappers for new bitwise ops and shifts.
decent_array/interoperability/_numpy/numpy_backend.py	Implements matrix_transpose, new in-place ops, floor-div/mod, and bitwise ops/shifts.
decent_array/interoperability/_jax/jax_backend.py	Implements matrix_transpose, new in-place ops, floor-div/mod, and bitwise ops/shifts.
decent_array/interoperability/_pytorch/pytorch_backend.py	Implements matrix_transpose, new in-place ops, floor-div/mod, and bitwise ops/shifts.
decent_array/interoperability/_tensorflow/tensorflow_backend.py	Implements matrix_transpose, new in-place ops, floor-div/mod, and bitwise ops/shifts.
decent_array/interoperability/__init__.py	Exposes newly added iop functions via imports and __all__.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Simpag]

Just one question and one change. The rest looks good

[Simpag]

What is this needed for?

[nicola-bastianello]

it's to handle the case of writing +x. not very common but it can happen

[Simpag]

fair enough, that would probably be more of a syntax error for this class but fine to keep

[Simpag]

This is a bit hacky and hard to extend. It would be cleaner if this is added to the iop functions. Maybe "dtype_of" or just "dtype" but the former is a bit clearer. Doing it in the iop layer would just require a reversed map of the one thats already in each backend, it would basically be something like:

_INV_DTYPE_MAP = {v: k for k, v in _DTYPE_MAP.items()}

def dtype_of(x):
    v = _INV_DTYPE_MAP.get(x.dtype)
    if not v:
        # Should never reach here since we control dtype but can be if user does something hacky
        raise SOME_ERROR
    return v

[nicola-bastianello]

yes this is definitely really hacky. I'm starting to work on a much more solid version now based on the ideas in #8. but that is beyond the scope of this PR, so I thought to leave this hacky placeholder for now

[Simpag]

Fair enough, feel free to merge when you feel ready